DynosAI
Public RC · Eval maturity

1.0 live cells stay not_run until real provider runs exist.

DynosAI 1.0.0-rc.3 still publishes MATRIX_1.0 with four provider-aware cells. None of them copy 0.13 Quality 100 scores. The historical table further down is the 0.13.0 core baseline, not 1.0 live proof.

MATRIX_1.0 live certification

ProviderModeStatus
CodexGreenfieldnot_run
CodexBrownfieldnot_run
CursorGreenfieldnot_run
CursorBrownfieldnot_run

Historical 0.13.0 core matrix

ProviderScenarioResultQualityOracle
CodexGreenfield · FibonacciPASS1008/8
CursorGreenfield · FibonacciPASS1008/8
CodexBrownfield · Contract discountsPASS10010/10
CursorBrownfield · Contract discountsPASS10010/10
63
MCP calls
0
MCP failures
0
scope requests
0
retries

What the matrix proves

Both supported providers can complete governed greenfield work with the current governed workflow.

Both can preserve and extend existing behavior in the brownfield scenario while passing independent Oracle checks.

Codex uses structured-primary MCP transport while Cursor uses the compatibility transport required by its CLI stream.

What it does not prove

It is not a universal guarantee across every language, repository size, CI system, security policy, or provider version.

DynosAI 1.0.0-rc.3 publishes MATRIX_1.0 placeholders plus eval-maturity evidence and still benefits from real-world feedback. It is not a production-ready 1.0 claim.

The Predictive Router remains in shadow mode even though the historical sample size now crosses its quantitative authority gates.